Sensitivity and Uncertainty Analysis of a Semi-Distributed Model used for the Quantitative Assessment of Watershed-Scale Hydrological Services of Wetlands

Wetlands have been recognized for their significant role on the hydrological cycle through non-linear relationships between storage and release of water. Despite this general agreement, since 1970 their spatial extent has diminished worldwide by 87%. To quantitatively assess their hydrological services at the watershed scale, specific wetland modules have been developed and integrated in distributed and semi-distributed models such as SWAT, MIKE SHE, SWIM, HydroGeoSphere, to name a few. For HYDROTEL, a physically-based semi-distributed model, two specific modules were developed to assess the services provided by isolated and riparian wetlands. Despite a local sensitivity analysis carried out on their parameters, five per wetland modules, too little attention has been paid to the overall sensitivity of these modules and other modules of HYDROTEL. Indeed, local methods are known for their incapacity to identify interactions between parameters, at the risk of erroneous sensitivity estimations, especially for models with large parametrization or highly non-linear characteristics. This is why we have conducted a systematic sensitivity analysis of all calibration parameters of HYDROTEL (snow - 8; soil - 7; interpolation - 3), and resulting uncertainty, with and without the wetland modules. This investigation was conducted for three watersheds of the St. Lawrence River with different proportions of wetlands, the Bécancour, Yamaska, and St. Charles River watersheds, Québec Canada. Sensitivity analyses were carried out using the SAFE toolbox in Matlab including seven methods among which Fourier amplitude sensitivity testing, GLUE, and PAWN, a moment-independent method. Whereas the uncertainties of several common hydrological data indices (2- and 10-yr 7-day and 5-yr 30-day low flows, and 2- and 20- yr high flows) were assessed using a Monte Carlo simulations in association with a meta model, GLUE, and DDS-AU. This study has provided a needed systematic framework to assess the impact of parameter sensitivity and uncertainty on HYDROTEL performances. It should prove useful to all future model applications intended to guide wetland policymakers and water management officials.